Opened 23 months ago

Last modified 10 days ago

#22729 new defect

Revisit relay read/write history resolution (for onion services)

Reported by: mikeperry Owned by:
Priority: Medium Milestone: Tor: unspecified
Component: Core Tor/Tor Version:
Severity: Normal Keywords: tor-hs, guard-discovery-stats, privcount
Cc: karsten Actual Points:
Parent ID: #22898 Points:
Reviewer: Sponsor: SponsorV-can

Description

At Wilmington, we wondered if we still were providing too fine-grained information in the read-write statistics, and if we should add noise there or reduce the reporting frequency.

It would be good to put bounds on the resolution, frequency, and noise we want here, as this would also help us decide what to do about things like #22422.

Child Tickets

Change History (8)

comment:1 Changed 20 months ago by mikeperry

See #23512 for an instance where these stats revealed too much information and aided in a deanonymization attack.

comment:2 Changed 19 months ago by mikeperry

Keywords: guard-discovery-stats added; guard-discovery removed

comment:3 Changed 19 months ago by mikeperry

Milestone: Tor: unspecified

comment:4 Changed 10 months ago by mikeperry

For reference, we did in fact raise our stats reporting interval to 24 hours for bandwidth stats. This ticket is about determining if that is enough.

comment:5 Changed 10 months ago by mikeperry

Component: Core Tor/TorDNSELCore Tor/Tor

comment:6 Changed 4 months ago by teor

Keywords: privcount added
Sponsor: SponsorV-can

We should think about this as part of PrivCount.

comment:7 Changed 4 months ago by teor

Parent ID: #22898

comment:8 Changed 10 days ago by mikeperry

In ticket:23512#comment:35, jaym reported that after the fix, write totals were larger than read totals for relays under attack by the oom side channel.

There are many possible sources for this discrepancy, and it may not actually not be visible under our current live network's stats resolution -- the discrepancy was shown only in chutney. We should see if we can reproduce it using his code at https://github.com/frochet/dropping_on_the_edge/tree/master/hs_drop_attack in a more realistic fashion

Note: See TracTickets for help on using tickets.